-
Notifications
You must be signed in to change notification settings - Fork 7
/
test_is_fid.py
149 lines (132 loc) · 5.11 KB
/
test_is_fid.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
import copy
import os
import pickle
import torch
import torch.nn as nn
import numpy as np
from library import inputs, eval_inception_score
from Utils.checkpoints import save_context, Logger
from Utils import flags
from Utils import config
import Torture
from library.chainer_evaluation.evaluation import calc_inception, calc_FID, FID
import PIL.Image
FLAGS = flags.FLAGS
KEY_ARGUMENTS = config.load_config(FLAGS.config_file)
dirname = model = FLAGS.old_model
config_path = os.path.join(dirname, "..", "source", "configs_dict.pkl")
summary_path = os.path.join(dirname, "..", "summary")
with open(config_path, "rb") as f:
new_dict = pickle.load(f)
new_dict["gpu"] = FLAGS.gpu
FLAGS.set_dict(new_dict)
FLAGS.old_model = "loaded"
text_logger, MODELS_FOLDER, SUMMARIES_FOLDER = save_context(__file__, KEY_ARGUMENTS)
torch.manual_seed(1234)
torch.cuda.manual_seed(1235)
np.random.seed(1236)
torch.backends.cudnn.deterministic = True
torch.backends.cudnn.benchmark = True
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
print(device)
itr = inputs.get_data_iter(batch_size=100, subset=1000)
itr_u = inputs.get_data_iter(batch_size=100, infinity=False)
netG, optim_G = inputs.get_generator_optimizer()
netD, optim_D = inputs.get_discriminator_optimizer()
netC, optim_c = inputs.get_classifier_optimizer()
netC_T, _ = inputs.get_classifier_optimizer()
netG, netD, netC = netG.to(device), netD.to(device), netC.to(device)
netG = nn.DataParallel(netG)
netC = nn.DataParallel(netC)
netC_T = nn.DataParallel(netC_T)
checkpoint_io = Torture.utils.checkpoint.CheckpointIO(checkpoint_dir=MODELS_FOLDER)
checkpoint_io.register_modules(
netG=netG, netC=netC, netC_T=netC_T, optim_G=optim_G, optim_c=optim_c,
)
logger = Logger(log_dir=SUMMARIES_FOLDER)
# dat1 = np.load("cifar10_total.npz")
# dat2 = np.load("cifar10_total_original.npz")
# fid = FID(dat1["mean"], dat1["cov"], dat2["mean"], dat2["cov"])
# print(fid)
# exit()
# x_list = []
# for _ in range(500):
# data, label = itr_u.__next__()
# x_list.append(data.data.cpu().numpy() * 0.5 + 0.5)
# x_list = np.concatenate(x_list) * 255
# print(np.min(x_list), np.max(x_list))
# print(x_list.shape)
# calc_FID(x_list)
# exit()
# x_list, y_list = [], []
# for _ in range(500):
# data, label = itr_u.__next__()
# x_list.append(data.data.cpu().numpy() * 0.5 + 0.5)
# y_list.append(label.data.cpu().numpy())
# x_list = np.concatenate(x_list) * 255
# y_list = np.concatenate(y_list)
# for lid in range(10):
# tmp_x = x_list[y_list == lid]
# calc_FID(tmp_x, 100, "./cifar10_label_{}.npz".format(lid))
# print(np.min(x_list), np.max(x_list))
# print(x_list.shape)
# exit()
# max_ism, max_id = 0, 0
# for iters in range(10000, 500001, 10000):
# torch.manual_seed(1235)
# torch.cuda.manual_seed(1235)
# np.random.seed(1235)
# model_path = os.path.join(dirname, "model{}.pt".format(iters))
# if os.path.exists(model_path) is False:
# continue
# checkpoint_io.load_file(model_path)
# # # # # Inception score
# with torch.no_grad():
# netG.eval()
# img_list = []
# for _ in range(10):
# sample_z = torch.randn(100, FLAGS.g_z_dim).to(device)
# data, label = itr.__next__()
# x_fake = netG(sample_z.to(device), label.to(device))
# img_list.append(x_fake.data.cpu().numpy() * 0.5 + 0.5)
# img_list = np.concatenate(img_list, axis=0) * 255
# img_list = (np.transpose(img_list, [0, 2, 3, 1])).astype(np.uint8)
# new_img_list = []
# for i in range(img_list.shape[0]):
# new_img_list.append(img_list[i])
# ism, isvar = eval_inception_score.get_inception_score(new_img_list, 1, 100)
# # ism, isvar = calc_inception(img_list, 100, 5000, 1)
# if ism > max_ism:
# max_ism = ism
# max_id = iters
# text_logger.info(str((iters, ism, max_id, max_ism)))
max_id = 420000
model_path = os.path.join(dirname, "model{}.pt".format(max_id))
checkpoint_io.load_file(model_path)
# # # # Inception score
with torch.no_grad():
torch.manual_seed(1235)
torch.cuda.manual_seed(1235)
np.random.seed(1235)
netG.eval()
img_list, label_list = [], []
for _ in range(50):
sample_z = torch.randn(100, FLAGS.g_z_dim).to(device)
data, label = itr.__next__()
x_fake = netG(sample_z.to(device), label.to(device))
img_list.append(x_fake.data.cpu().numpy() * 0.5 + 0.5)
label_list.append(label.data.cpu().numpy())
img_list = np.concatenate(img_list, axis=0) * 255
label_list = np.concatenate(label_list)
# for iid, img in enumerate(img_list):
# img = PIL.Image.fromarray(img.transpose([1, 2, 0]).astype(np.uint8))
# img.save("/home/kunxu/cifar10_tmp/{}.png".format(iid))
# img_list = np.transpose(img_list, [0, 2, 3, 1])
# print(img_list.shape)
# new_img_list = []
# for i in range(img_list.shape[0]):
# new_img_list.append(img_list[i].astype(np.uint8))
ism, isvar = calc_inception(img_list, 100, 5000, 1)
text_logger.info(str((max_id, ism, isvar)))
fid = calc_FID(img_list, 100, "./cifar10_total.npz")
text_logger.info(str((max_id, fid)))